| { | |
| "blocks": 16, | |
| "compile": false, | |
| "configuration_kind": "Recipe summary; complete model argument and module/distributed initialization must follow the pinned trainer.", | |
| "curvature_c": 1.0, | |
| "effective_batch_sequences": 64, | |
| "eos_token_id": 128001, | |
| "eval_every": 5000, | |
| "final_learning_rate": 0.0001, | |
| "helm_revision": "7501deca8f413848bfef804be64ce874b72a3cd7", | |
| "learning_rate": 0.001, | |
| "max_sequence_length": 2048, | |
| "model_class": "LorentzDeepSeekV3", | |
| "parameter_precision": "fp32", | |
| "parameters": 966957719, | |
| "planned_steps": 152580, | |
| "quantized_attention": false, | |
| "tokenizer": "NousResearch/Meta-Llama-3-8B", | |
| "trainable_curvature": false, | |
| "trainer_revision": "5f968bd8d023298728e3ac43d6fefe64f786de8e", | |
| "training_precision": "bf16", | |
| "warmup_steps": 3051, | |
| "width": 910 | |
| } | |